4.1.

X-Ray Sources and Beam Control

One of the key values in x-ray-based molecular sensing is the concept that the position of the excitation source is known a priori and can be used in an image reconstruction process, such that even if the emission is blurred by optical scattering through tissue, there can be a high fidelity image recovered. This was initially demonstrated in x-ray optical luminescence tomography by Pratx et al.,³ using keV x-ray imaging and image reconstruction. Diffusion modeling can be employed to reduce errors associated with the intensity of the luminescence emission being reduced by the tissue propagation. The geometries used directly follow those available for x-ray CT because the source technology is derived from this. In the keV energy range, swept line scan beams or partial angle tomography is possible,^12,29,32 and in MV photon beams from linear accelerators, multi-leaf collimators (MLC) and jaws are available to shape the beams dynamically.³³ Full field imaging can work as well, albeit without significant axial resolution because of optical scatter in the tissue, but providing apparent high resolution in lateral imaging of the tissue surface.³⁴

4.2.

Temporal Acquisition

The first studies of x-ray-induced optical luminescence based upon scintillation did not acquire in temporal sampling because the x-ray sources were continuous, and so there was no inherent value to temporal sampling. However, in Cherenkov-excited luminescence imaging, there is both inherent value in removal of the background Cherenkov excitation light, as well as the benefit of lock-in detection to the pulsed x-ray source. Linear accelerators produced for clinical use most commonly have pulsed electron bunches that are accelerated in short 3- to 5-ms pulses, with a low repetition rate near 100 to 400 Hz. This sampling is significantly slower than used for time-resolved fluorescence but fits with time-resolved phosphorescence methods, and so the first in vivo demonstrations have focused on imaging luminescence from oxygen sensors that have triplet states that are quenched on this timescale. Faster pulsed x-ray sources exist though, such as small x-ray sources for portable x-rays²⁰ or even large sources used for industrial pulsed radiography.³⁵ The fastest commercially available pulsed x-ray sources are in the tens of nanoseconds range for pulse length, and the completeness of the fall off is usually uncharacterized or unspecified. Thus, fast time gating for things such as fluorescence from organic molecules, with lifetimes usually in the nanosecond range, would need a deconvolution for the x-ray-detector instrument response function,^36,37 as is commonly done in time-correlated single photon counting work.³⁸

Temporal gating has been demonstrated with luminescence emission to suppress the Cherenkov light signal and allow for a nearly background-free sensing of the tissue. Single lymph node imaging was demonstrated by Zhang et al,³³ using an oxygen sensing agent, and the lifetime recovery between 22 and 44 ms provided the ability to sense the local partial pressure of oxygen (${\mathrm{pO}}_2$). Sensing of other luminescent species is readily possible and Europium microspheres are commercially available agents for binding with targeting moieties that have a luminescence lifetime in the 100’s of microseconds.³⁹ Alternatively, silicone nanoparticles also have a long lived lifetime and can be used as a light signal generator with targeted delivery.⁴⁰

4.3.

Lateral and Axial Spatial Resolution

Perhaps one of the most undeveloped areas in x-ray-based molecular sensing is control of the x-ray source for improvement in spatial resolution. Advances in conformal and adaptive radiotherapy have led to improved tools on a Linac to control beam dose deposition. The MLCs present on the output of the Linac have advanced to a very high degree of precision, where millimeter level accuracy in dose fall off can be achieved. This same precision can then be deployed where the lateral and axial extent of the beams are used to adjust the sampling of tissue. As mentioned above, lateral resolution is largely controlled by the MLCs and jaws of the Linac, but the axial resolution is determined by choices of the beam radiation type (electrons, photons, and protons) or beam depth of scanning, as illustrated in Fig. 4(a). Thus, the Linac MLCs provide a simple technological way to shape the beams for line scans, point scans, multi-point or multi-line scans [see Fig. 4(a)], or even more generally a set of orthogonal basis functions in the source patterns for approaches such as compressed sensing.^41,43,44

Fig. 4

The lateral spatial confines of the beam in tissue are determined by the Linac (a) beam shaping by the MLCs (photo inset below) with illustration of how the MLCs can be used for point raster scanning or multipoint scanning.⁴¹ (b) The depth of sensing is affected by the choice of radiation and beam energy, as illustrated by depth–dose curves for electron beams. The scan could be axially modulated in depth by variation of buildup bolus between the tissue and the beam, as illustrated in (c). Lateral resolution is limited by the ability to slice or focus the x-ray beam, as illustrated by Monte Carlo simulations of at 6- and 18-MV beams (d), where the beam is directed downward and the $X\text{-}Y$ axes show the mm dimensions into the tissue and the color bar is a Cherenkov intensity scale.⁴² The final lateral spatial resolution can be deconvolved with the beam width, show sensitivity to sub-millimeter objects, as illustrated in (e) where thin luminescent capillaries of varying diameter, $d$, were used to show sensitivity to resolve the full width at half maximum (FWHM) down to 0.2-mm diameter.³³ The FWHM is a convolution of the beam width with $d$.

If the depth of penetration is to be restricted, it is feasible to use electron irradiation which has limited depth of penetration, and even apply a variable bolus or buildup region to the beam before the tissue, illustrated in Fig. 4(b), thereby allowing the depth of sampling to be varied externally to the Linac, just like an MLC adjusts the lateral component of the beam. The depth of sampling with electrons could be varied as they have the steepest fall off curve as compared to photons. Alternatively, in keV systems, various attempts have been shown to use x-ray focusing mirrors or fiber optic tapers^45–47 and as this technology evolves it might be feasible to limit lateral resolution with this focusing approach.

4.4.

Multiplexing of Signals

Excitation with x-rays implicitly leads to a broadband signal that can excite more than one molecular probe. Both scintillation and Cherenkov-excited signals can occur, and it is possible to detect multiple emission signals, based either on temporal gating or wavelength separation of intensity. The excitation of multiple molecular dyes by Cherenkov has been shown, with optimal spectral windows being in the red,^48,49 near-infrared, and short-wave infrared (SWIR) (also referred to as near-infrared window II, NIR-II)^50–52 wavelength bands. In particular, SWIR emitting agents are often nanoparticulate and thereby optimally suited for direct x-ray excitation.^50,53,54 There is a high volume of research in this area, and multicolor emitters are available.^55,56 Detectors for SWIR wavelengths are often different than those for the red or NIR wavelengths,⁵⁷ and so it is feasible to use separated detectors for parallelization or spectral decomposition through a spectrometer.^58,59 Time-gated detection can be coupled with wavelength-based separation of the signals as well, to further maximize the ability to detect signals at the same time, and scintillation-based signals can be orders of magnitude faster than organic fluorescence, which in turn is orders of magnitude faster than phosphorescent signals. Careful sampling of the time-sequence of the x-ray pulses with the detector gating can optimize this.

5.1.

X-Ray Beam Location Used as Prior Spatial Information in Imaging

X-ray photons experience much less scattering than optical radiation in tissue, and therefore x-ray-induced radioluminescence largely originates within or very near the volume that is directly in the pathway of the scanning beam. The limits on resolution though can be defined by the x-ray Compton scattering, which can happen both inside and outside the tissue. For keV sources, the scattering is usually not significant because it largely produces soft x-rays that have very short penetration depth in tissue. This is true both outside the tissue and within the tissue. In the MeV approach, the Linac MLC (Fig. 4) provides a simple technological way to shape the beams for various scans as mentioned above, but also to induce x-ray scattering near the leaf boundaries, to some degree, depending on the specific leaf end structure.^60,61 A similar situation exists within the tissue, where in Cherenkov-excited luminescence emission, most Cherenkov photons (in the UV-blue spectral range) have very short diffusion path ($<1\mathrm{mm}$) in the tissue due to the hemoglobin and water absorption. These properties constitute the key aspect of the x-ray-induced molecular imaging scheme, where the distribution of optical signal along the direction of scanning can be recovered by measuring total luminescence signal and considering that the signal all originated from the position of the scan beam in the tissue, used as prior information in a reconstruction algorithm.

5.2.

Diffusion Modeling of Cherenkov and Optical Photons

The radiative transfer of photon diffusion through turbid media can be solved either numerically through Monte Carlo simulation or approximated analytically by the diffusion approximation. Monte Carlo solutions generally maintain high accuracy and wide applicability,^62,63 and their use has been tremendously accelerated by computation on GPUs.^64–69 There are extensions of this to model high-energy particle transport that can also be used, e.g., GEANT4, although with considerably more computational effort.^70–73 The diffusion approximation provides a first-order approximation of transport over distances beyond a few millimeters, and it has been broadly applied in diffuse optical tomography. The limits to this approximation though are important, meaning that it must be applied only where the detected photons are scattered sufficiently to lose their original directions. This condition could be marginally met in Cherenkov imaging depending in the red and NIR wavelength bands.^74,75 For example, the emission UV-blue spectral range that always predominates the measurement is significantly absorbed in most biological tissues, and so is rarely diffuse in transport.^76,77

Cherenkov light is generally modeled in a comprehensive manner that involves distinct behavior of a broadband wavelength spectrum of photons peaked in the ultraviolet and decaying in intensity with an inverse square dependence upon the wavelength.⁷⁸ Modeling of x-ray-induced Cherenkov as the excited light contains two coupled processes of excitation and emission, and explicitly would require modeling the x-ray interactions with Monte Carlo tools such as GEANT4. In practice, the Cherenkov/x-ray beam can be assumed to be non-diffuse source of light inside the tissue, to simply the process. The Monte Carlo codes that model radiation-induced light transport in biological media have been integrated into the GAMOS interface to GEANT4.⁷³

5.3.

Typical Image Reconstructions: Advantages and Limitations

Currently, image reconstructions could be coarsely classified into three categories: model-based^79,80 direct deconvolution,^43,81,82 and back-projection.^32,83 The model-based approaches have the widest applicability to measurement geometry and excitation mechanisms. However, a model-based reconstruction can be inefficient and even ineffective when the target is close to or far from the boundary at the detection side. In the former case, a stringent forward model using either Monte Carlo or higher-accuracy approximations to exact radiation transport modeling is requisite; while in the latter case, any tiny perturbation and inconsistency of the optical properties could lead to erroneous reconstructions on both the intensity and location. The recontruction can be improved by combining useful prior information about the measurment geometry, charateristic emissions, etc., into the regularization process, which has been an effective way, albeit with the assumption of the priors being accurate.^84–90

By taking advantage of a careful measurement geometry and utilization of coded illumination techniques, a straightforward direct deconvolution has been used for reconstruction, e.g., Cherenkov-excited luminescence scanned imaging (CELSI). CELSI uses the collimating system [Fig. 4(a)] of a Linac to send a sheet of radiation traveling across the imaged subject in a manner equivalent to the excitation-beam shaping used in light-sheet microscopy, as illustrated in the first row of Fig. 5. By restricting the excitation beam to a single, narrow sheet, the origin of the optical photons can be inferred regardless of where these photons were detected or how many times they scattered in tissue. Direct deconvolution was also applied for a vertical illumination geometry as shown in the second row of Fig. 5. By simultaneously capturing both Cherenkov and CELSI images that were excited via spatially modulated x-ray beam, the image recovery was improved by a spatial demodulation strategy at each time step based upon compressed sensing techniques. However, direct deconvolution typically suffers from two issues: (i) accurate convolutional kernels are generally hard to be determined with acceptable computational cost and (ii) numerical deconvolution could magnify the measurement noise when the amplitude approaches zero in the frequency domain.

Fig. 5

Cherenkov radioluminescence measurement geometries based on Linac MLC delivery and corresponding non-model-based image reconstructions. Laterally scanned imaging (top) sends sheet-shaped x-ray beams to measure local luminescence emission that is then deconvolved with the kernels derived from a diffusion approximation. Vertically scanned imaging (middle) spatially and temporally modulates the Cherenkov illumination following specific beam patterns and then demodulation with the simultaneously acquired excitation Cherenkov beam shapes is possible. Like the imaging technique used in CT (bottom), CELSI images were acquired by rotating the Linac gantry to yield a sinogram, which was then used with FBP of the data for image reconstruction. In all experiments, a plane mirror was always used to avoid direct exposure of the imaging camera to the radiation beam.

Backprojection or filtered backprojection (FBP) reconstruction algorithm originally developed for CT has been used, known as x-ray luminescence CT (XLCT).^3,91 Like the measurement geometry in CT, raw images in XLCT were acquired by rotating the Linac gantry (see the last row of Fig. 5) to render a sinogram that was used in backprojection-based image reconstruction. Each element in the sinogram is an integration through the pixels of a captured image, essentially implementing single-pixel imaging that facilitates the use of single-photon point detector for high-sensitivity measurement in conjunction with the TCSPC technique and the use of a spectrometer for hyperspectral studies. In XLCT, anatomical prior information can be provided by either x-ray imaging or Cherenkov surface data.^44,92–94 Since the measurements are taken from multiple views, XLCT in-vivo imaging is similar to CT imaging in that the image is always better with more radiation, which can lead to excessive radiation dose, and so care must be taken to design the scan sequence. To address this issue, a lot of effort has focused on sparse-view XLCT by taking advantage of improved measurement geometry and algorithms.^92,95,96

7.1.

Molecular Sensing of Oxygen and pH

Tumor oxygenation significantly affects the outcome of radiotherapy, and hypoxic tissues such as tumors (defined as having partial pressure of oxygen, ${\mathrm{pO}}_2$, less than 10 mmHg) are known to be more resistant to radiation damage than fully oxygenated tissue.^127–129 Therefore, monitoring of tumor oxygenation is thought to be highly desirable for effective radiotherapy. Cherenkov-based imaging provides an internal light source to excite oxygen-sensitive phosphorescence probes during radiotherapy (Fig. 6).³⁴ Tomographic oxygen images were reconstructed in experimental mice, to sense the ${\mathrm{pO}}_2$.^118,119 Spatial resolution was improved by spatially encoding of the beam positioning with the Linac MLCs and subsequent deconvolution of the beam width from the signal.^120,130 Using a scanning sheet-shaped MV x-ray beam, the excitation was restricted to narrow volumes, and tomographic oxygenation imaging with sub-millimeter resolution and nanomolar sensitivity at depths of several millimeters was demonstrated (Fig. 6).⁸⁰ An alternative approach used the MV x-ray-induced Cherenkov emission spectral changes from differences in attenuation by the blood oxygen saturation, ${\mathrm{SO}}_2$ values. Tissue blood ${\mathrm{SO}}_2$ alters the absorption of the emitted Cherenkov, by de-oxygenated blood absorbing more in the 600- to 750-nm wavelength band. The change in spectrum could be sampled, or more simply the change in broadband intensity could be tracked during radiation delivery.^131,132

Fig. 6

Cherenkov-excited luminescence imaging of tumor oxygen. (a) Structure and oxygen quenching property of probe PtG4.³⁴ (b) In vivo ${\mathrm{pO}}_2$ imaging of tumor and normal tissue of a mouse after IV injection of PtG4.³⁴ (c) Maximum-intensity projection luminescence image overlaid to x-ray CT scan and tomographic luminescence of PtG4.⁸⁰

Another typical tumor microenvironment is the acidosis induced by high metabolic rate in poorly perfused regions of tumor, as a result extracellular pH in tumors is lower than in normal tissue and can be correlated with prognosis and response to treatment.^133,134 An x-ray-excited pH sensor consisted of an x-ray scintillator film (${\mathrm{Gd}}_2{\mathrm{O}}_2\mathrm{S}:\mathrm{Tb}$ and ${\mathrm{Gd}}_2{\mathrm{O}}_2\mathrm{S}:\mathrm{Eu}$) coated in a methyl red-dyed paper, used for high-resolution pH detection in tissue. The pH was obtained by analyzing the optical spectrum through the paper after scanning with narrow x-ray beam.¹³⁵ Then the noninvasive pH sensing was used to monitor bacterial infection and treatment of implanted medical devices through tissues after coating on implanted surfaces.^110,111 Using a pH-triggered nanocapsule, spectral changes were sensitive to the release of doxorubicin, which can be used to track drug release in the acidic tumor microenvironment.¹⁰⁹ Some pH-dependent long-lived emission luminescent probes appear promising for Cherenkov-based lifetime detection.^136–138

7.2.

Sensing of Immunology: Biological Imaging and Development of Combined Therapeutics

In the past few decades, the innovation of immunotherapy has revolutionized the approach to treatment of advanced cancer by means of immune cell therapy, stimulating the immune system to destroy malignant cells.^139–141 Studies have used fluorescent imaging of NK cells within human prostate cancer xenografts¹⁴² and migration of T-cells to tumors.^143,144 This fluorescent labeling technique for immune cells could be also used with x-rays to perform in vivo real-time monitoring of therapeutic effects for immune cell-based therapy. The relevance of this is because the combination of radiotherapy and immunotherapy is now shown to enhance the induced systemic anti-tumor response and achieve higher tumor control effect.¹⁴⁵ Imaging has the potential to monitor immune signals during fractionated radiotherapy for patient specific sensing of the synergy between these treatments and their timing.

7.3.

Theranostics: Molecular Sensing with X-Ray-Induced Photodynamic Therapy

The use of light as an activation mechanism has a long history in therapeutics, perhaps the most common being UV treatment for psoriasis^146–149 and photodynamic therapy (PDT).^150–152 PDT is a noninvasive cancer treatment modality that utilizes photosensitizers to generate cytotoxic reactive oxygen species when activated by light of appropriate wavelengths.^153,154 PDT has its core strength in the shallow tissue penetration of illumination light, and this is widely utilized in dermatology for superficial skin lesion treatment.^155–158 However, treatment of deep lesions has been limited by blood attenuation of the light.^159–161 Using x-rays as the irradiation source for PDT is a way to overcome this problem. This has been extensively studied in recent years, with many approaches using nanoscintillators to convert the x-rays to UV/visible light, which activates photosensitizer deep in tissues,^162,163 and examples with direct Cherenkov activation are possible.

This approach to x-ray-activated photodynamic therapy (XPDT) makes it feasible to integrate diagnosis and tumor therapy for tumor theranostic applications. A theranostic mesoporous silica nanoparticle encapsulated a photosensitizer, 2,3-naphthalocyanine and a ${\mathrm{LiGa}}_5{\mathrm{O}}_8:\mathrm{Cr}$-based nanoscintillator was designed to efficiently mediate deep tissue XPDT and guide the irradiation by x-ray-excited luminescence imaging.¹⁶⁴ This nanoparticle still produces ${}^1{\mathrm{O}}_2$ from the long and intense afterglow luminescence of ${\mathrm{LiGa}}_5{\mathrm{O}}_8:\mathrm{Cr}$ after x-ray irradiation. After conjugation with Cetuximab (i.e., antibody to the EGF receptor), the nanoparticles were able to selectively accumulate in EGFR expressing orthotopic lung tumors for both EGFR-mediated molecular sensing with x-rays and XPDT. Another example constructed an x-ray-excited core–shell–shell theranostic scintillator nanoparticle based on lanthanides-doped ${\mathrm{NaLuF}}_4$, which could emit visible light, and was used with rose Bengal photosensitizer for PDT and imaging.¹⁰⁰ Alternative approaches to drug release have used poly(lactide-co-glycolide) polymeric nanoparticles incorporating a photosensitizer (verteporfin) that could be triggered by 6-MeV x-rays to generate ${}^1{\mathrm{O}}_2$. In addition, targeting of nanoparticles with folic acid enables specific targeting of tumors that overexpress the folate receptors. Inclusion of radiation-activated ${\mathrm{TiO}}_2$ nanoparticles has been shown to have therapeutic effect, postulated to be mediated by Cherenkov light causing a photocatalytic effect, leading to radical species production.¹⁶⁵

X-ray deposited chemotherapies have been examined in several delivery moieties. Delivery via vitamin B12 uptake via the transcobalamin receptor was shown for delivery of agents and photorelease via the alkylcobalamin scaffold that is light sensitive.¹⁶⁶ X-ray activation was shown to release doxorubicin bound in micelles, by breakage an o-nitrobenzyl linker, thereby breaking open micelles, and resulting in delivery of doxorubicin to the nucleus.¹⁶⁷

While this last section has focused on therapeutic effects from x-rays mediated by optical signals, there is a linkage to diagnostic scanning that could be very important, both scientifically and practically. As these therapeutic studies advance, it seems inevitable that the diagnostic potential for their use will become more apparent as well.